Running head: VISUAL SIMILARITY IN THE MOSES ILLUSION Here's Looking at You: Visual Similarity Exacerbates the Moses Illusion for Semantically Similar Celebrities Danielle K. Davis & Lise Abrams University of Florida Author Note Danielle K. Davis, Department of Psychology, University of Florida; Lise Abrams, Department of Psychology, University of Florida. We thank our undergraduate research assistants Lea Martin, Maddie Gonzalez, Stefani Pila, Katrina Gaffney, Jessica Campbell, and Jack Mizell for their assistance with stimuli development, program testing, and data collection. We also thank Edith Kaan, Andreas Keil, and Katherine White, for their comments on an earlier version of this manuscript. We gratefully acknowledge IAU College in Aix-en-Provence, France, for providing resources to Lise Abrams during her sabbatical as a resident fellow in 2013-2014. This research was presented at the 55th annual meeting of the Psychonomic Society and is based on Danielle K. Davis's master's thesis. Correspondence concerning this article should be sent to: Danielle Davis, Department of Psychology, University of Florida, P.O. Box 112250, Gainesville, FL 32611-2250, E-mail: [email protected]; or Lise Abrams, Department of Psychology, University of Florida, P.O. Box 112250, Gainesville, FL, 32611-2250, E-mail: [email protected] VISUAL SIMILARITY IN THE MOSES ILLUSION 2 Abstract When people read questions like How many animals of each kind did Moses take on the ark?, many mistakenly answer "two" despite knowing that Noah sailed the ark. This "Moses illusion" occurs when names share semantic features. Two experiments examined whether shared visual concepts (facial features) exacerbate Moses illusions for celebrity names. Questions contained an unrelated distractor name or a semantic distractor name that was visually similar or dissimilar to the correct target name. Both experiments revealed more Moses illusions occurred for questions containing a visually similar semantic distractor compared to either visually dissimilar or unrelated distractors. Furthermore, presenting a picture of the target (Experiment 1) or the visually similar distractor (Experiment 2) before the question increased accurate detection of the illusion, independent of distractor type. Results challenge theoretical explanations of the Moses illusion as resulting from purely shallow semantic processing and demonstrate the importance of visual information in processing proper names, even when presented in written form. Keywords: name recognition, facial features, Moses Illusion, reading, language VISUAL SIMILARITY IN THE MOSES ILLUSION 3 Here's Looking at You: Visual Similarity Exacerbates the Moses Illusion for Semantically Similar Celebrities Proper names are particularly difficult to learn and remember compared to other types of words or biographical information (Barresi, Obler, & Goodglass, 1998; Fraas et al., 2002; James, 2004; Stanhope & Cohen, 1993; Young, Hay, & Ellis, 1985). Most research has focused on names' decreased retrieval (i.e., fewer successful retrievals and more tip-of-the-tongue states; e.g., Burke, MacKay, Worthley, & Wade, 1991; Hanley & Chapman, 2008), as it is considerably easier to recognize names (e.g., Peressotti, Cubelli, & Job, 2003). However, under some circumstances, name recognition during comprehension can become more difficult. One example is the Moses illusion, a recognition error that occurs in response to questions like How many animals of each kind did Moses take on the ark? Many people answer “two”, only realizing their failure to catch the mistake after being reminded that Noah, not Moses, took animals on the ark (e.g., Erickson & Mattson, 1981). Moses illusions have been proposed to result from shallow semantic processing of the question’s meaning, where not every word's meaning is thoroughly checked for congruity (e.g., Erickson & Mattson, 1981; Kamas, Reder, & Ayers, 1996; Reder & Kusbit, 1991; Song & Schwarz, 2008). Consequently, a distractor name similar to the meaning of the question (e.g., a biblical figure, Moses) is less easily detected as anomalous than a dissimilar name (e.g., Nixon). Furthermore, more illusions occur when the distractor and target names have high semantic overlap (e.g., Moses and Noah) compared to low (e.g., Adam and Noah; van Jaarsveld, Dijkstra, & Hermans, 1997; van Oostendorp & de Mul, 1990). However, people can share not only semantic information (e.g., both Al Gore and Joe Biden are politicians, were vice presidents, etc.), but also visual features (e.g., both are older males and have graying hair), the latter of which are critical for recognizing and distinguishing people. This experiment explored VISUAL SIMILARITY IN THE MOSES ILLUSION 4 whether visual concepts, specifically facial features, play a significant role in inducing the Moses illusion. Support for non-semantic influences on the Moses illusion comes from Shafto and MacKay (2000, 2010). Shafto and MacKay (2000) demonstrated that phonological overlap could induce the illusion independently from semantics, as more illusions occurred when replacing the name Neil Armstrong (astronaut) with Louis Armstrong (musician) compared to Dizzy Gillespie (musician), even though Louis Armstrong and Dizzy Gillespie were equally semantically distinct from Neil Armstrong. Although this finding can be challenged because both Neil and Louis shared a lexical-surname node (meaning that Armstrong was still semantically compatible with the sentence), they demonstrated additional evidence in support of phonological influences by showing greater susceptibility to the illusion when semantic distractors were also phonologically related to targets (e.g., Lyndon Johnson and Andrew Johnson). Shafto and MacKay (2010) also showed a pure phonological effect using lexically-distinct homophones, i.e., more illusions occurred when mail replaced male in a sentence compared to mill or shed. These results were problematic for a purely semantic explanation of the Moses illusion, and they instead proposed an account within Node Structure Theory (NST), an interactive activation model of lexical access (e.g., MacKay, 1987), where names are represented in multilevel systems (see Figure 1). Consider the person Brad Pitt: Within NST, propositional nodes (biographical information; e.g., actor) and visual concept nodes (a holistic representation of what the person looks like, also connecting to visualizable propositions like "has blue eyes") independently connect to lexical nodes (the first name Brad and last name Pitt, as well as the proper name phrase Brad Pitt), which then connect to phonological nodes (the names' sounds). Within NST, the Moses illusion occurs when the target (the correct agent of the sentence) VISUAL SIMILARITY IN THE MOSES ILLUSION 5 receives a convergence of priming from multiple sources, resulting in its activation instead of the distractor. Consider the earlier example, where Moses is the distractor and Noah is the target. While the distractor's lexical node receives bottom-up priming when the written word "Moses" is perceived, the target's lexical node receives top-down priming via propositional nodes corresponding to information listed in the question as well as priming from the distractor via shared semantic links (e.g., both are biblical leaders). Thus, instead of activating the distractor (which would allow people to detect that the name did not fit the question context), the non- presented target receives the most priming, its lexical node becomes activated, and the distractor is miscomprehended as the correct name for that context, resulting in the illusion. Experiment 1 Theoretical explanations where converging sources of priming cause the Moses illusion allow for the possibility of shared facial features between the target and distractor to exacerbate the illusion, as illustrated in Figure 2. In NST, facial features are housed in “visual concept nodes” that also connect to visualizable propositions. These nodes are connected to other propositional nodes at the semantic level as well as the lexical node (MacKay & Burke, 1991) and become activated when recognizing an image of the person or even discussing a person. Thus, visual information is accessed even when the task does not explicitly require identification of visual characteristics (e.g., while reading biographical information). To test the hypothesis that visual similarity is relevant for inducing the Moses illusion, Experiment 1 used celebrity names, as these are people whose facial features are familiar to college students. Specifically, we predicted that when a question contains a distractor (e.g., Chris Hemsworth) that is both semantically related and visually similar in terms of facial features shared with the target (e.g., Brad Pitt), the target receives additional priming from the shared connections between visual VISUAL SIMILARITY IN THE MOSES ILLUSION 6 concept nodes, which should result in greater susceptibility to the illusion relative to a visually dissimilar semantic distractor (e.g., Hugh Jackman). Furthermore, if visual similarity between a target and a distractor increases susceptibility to the Moses illusion, then highlighting visual differences between the target and the distractor should reduce susceptibility. Reder and Kusbit (1991; see also Kamas et al., 1996) observed that presenting relevant facts about the targets in text form (e.g., Noah took two animals of each kind on the ark) before reading invalid questions increased participants' awareness of anomalous distractors.